Battery cabinet maintenance method and system based on failure prediction and health management

Abstract

The application relates to the technical field of energy storage battery management, and discloses a battery cabinet maintenance method and system based on fault prediction and health management. In the method, multi-dimensional operation parameters of a battery cabinet are collected and stored in classification according to operation stages, a dynamic weight and a deviation degree are calculated to obtain a staged health index, a single-body health degree score is obtained by fusing the index, a health score is corrected in combination with a historical overrun frequency, a deviation value of a single body and a group median is accumulated to obtain a historical deviation total value, a dynamic boundary is determined according to historical parameters, a comprehensive risk score is obtained by comparison, a time window is divided to extract a characteristic vector and calculate a trajectory deviation degree, and abnormal transition is identified. A group deviation index is calculated by taking voltage, temperature and SOC as the calculation basis to identify an outlier subset, a capacity efficiency decay index is obtained by counting a capacity decay rate and a charging and discharging efficiency, a maintenance urgency index is obtained by weighting multiple indexes, and a graded maintenance decision is output. The application realizes multi-dimensional prediction and active maintenance, improves safety and delays cell aging.

Description

Technical Field

[0001] This application relates to the field of energy storage battery management technology, and more specifically, to a battery cabinet maintenance method and system based on fault prediction and health management. Background Technology

[0002] Battery cabinet health status assessment is a crucial foundation for formulating operation and maintenance strategies for energy storage systems. Current technologies generally employ single-parameter threshold judgment methods to identify battery anomalies, relying solely on individual cell voltage differences for status assessment. However, battery cabinets are complex electrochemical systems, and their operating status is influenced by a combination of factors, including temperature distribution, depth of charge / discharge, cycle count, and internal resistance changes. Traditional solutions lack systematic fault prediction and comprehensive health management methods. Relying on a single dimension for assessment easily leads to missed status diagnoses, failing to accurately reflect the equipment's health status. Battery cabinets with performance degradation that doesn't trigger alarms operate with defects for extended periods, accelerating cell aging. Furthermore, existing maintenance models are outdated and reactive, hindering refined and preventative maintenance. Summary of the Invention

[0003] The main purpose of this application is to provide a battery cabinet maintenance method and system based on fault prediction and health management, which aims to solve the technical problems of traditional battery cabinets relying on a single parameter threshold for judgment, lacking fault prediction and health management, resulting in serious missed status judgments, passive and delayed maintenance, and accelerated battery aging.

[0004] The first aspect of this application proposes a battery cabinet maintenance method based on fault prediction and health management, including: Collect multidimensional operating parameters of the battery cabinet and store them according to the operating stage. Calculate the parameter variation coefficient and dynamic weight, and obtain the staged health index by weighted summation of parameter deviation. The individual health score is obtained by fusing the phased health index, and the corrected health score is calculated by combining the historical out-of-limit frequency. The total historical deviation value is obtained by accumulating the deviation difference between the individual and the group median. The dynamic boundary is determined based on historical parameters. The current parameters are compared with the dynamic boundary to obtain a comprehensive risk score. The time window is divided to extract feature vectors to calculate the trajectory deviation. Abnormal transitions are identified by combining the parameter change rate. The population deviation index is calculated using voltage, temperature and SOC as dimensions to identify outlier subsets, and the capacity efficiency degradation index is obtained by statistically analyzing the capacity decay rate and charge / discharge efficiency. The maintenance urgency index is obtained by weighted summing of the historical deviation total, comprehensive risk score, trajectory deviation degree, abnormal jump and outlier subset proportion. Based on the maintenance urgency index and capacity efficiency decay index, a graded maintenance decision is output.

[0005] Furthermore, the steps of collecting multidimensional operating parameters of the battery cabinet and storing them according to operating stages, calculating the parameter variation coefficient and dynamic weight, and obtaining a staged health index by weighted summation of parameter deviations include: Collect multi-dimensional operating parameters of the battery cabinet, divide the time sequence into different operating stages and form a stage parameter matrix, fill in missing data, and complete parameter classification and storage. The coefficient of variation is calculated for the parameter matrix of each stage. The weighted coefficient of variation is obtained by introducing the time decay factor. The comprehensive coefficient of variation is extracted and normalized. The dynamic weight is determined after secondary correction and weight compensation. The parameter deviation is calculated based on the stage statistical benchmark, and the deviation is truncated and fused. The final deviation is obtained by combining the parameter change information, and the abnormal frequency is fed back to the weight iteration. Multiply the dynamic weights by the final absolute value of the deviation and sum them up to obtain the phased health index for the current stage. Store the health indexes by stage and establish a time series sequence. Output the health index values ​​with stage labels.

[0006] Further, the steps of calculating the coefficient of variation for the parameter matrix at each stage, introducing a time decay factor to obtain a weighted coefficient of variation, extracting and normalizing the comprehensive coefficient of variation, and determining the dynamic weights after secondary correction and weight compensation include: Read the parameter matrix of each stage and sort it by time, set the corresponding time window, filter the valid data segment and remove abnormal data, and organize it into a standard parameter sequence according to the timestamp. The basic coefficient of variation is calculated for the standard parameter sequence, the time decay coefficient is assigned and the weighted statistic is calculated to obtain the weighted coefficient of variation, and the parameter covariance matrix is ​​extracted to obtain the comprehensive variability index. The statistical parameters are used to generate a comprehensive correction coefficient based on the abnormal correlation characteristics. The weighted coefficient of variation is then double-corrected by combining the degree of variation index. Finally, a lower limit constraint on the weights is applied to obtain the intermediate weight value. The intermediate weight values ​​are normalized, and the abnormally frequent parameters are compensated. The compensated weights are normalized again, and the final dynamic weights are output and the weight iteration records are stored.

[0007] Further, the step of obtaining an individual health score based on the fusion of the phased health index, calculating a corrected health score by combining the historical exceedance frequency, and accumulating the deviation difference between the individual and the population median to obtain the total historical deviation value includes: Read the health index of each operation stage and perform the corresponding numerical transformation. Assign weights according to the operation stage and perform weighted fusion. After removing abnormal extreme values, obtain the individual health score. The historical parameters of individual units are statistically analyzed to exceed the limits. The frequency of each item exceeding the limit is calculated by hierarchical weighting and introducing a time decay factor. The weighted fusion is used to obtain the comprehensive frequency of exceeding the limits, and the corrected health score is obtained by calculating it with the individual health score. Extract and sort the population modified health scores to determine the population median, calculate the absolute difference between the individual and the median, apply an adjustment coefficient based on the historical deviation level, and obtain the adjusted deviation difference; The adjusted deviation difference is added to the historical cumulative deviation value, weighted according to time and accumulated. Constraint processing is applied to the cumulative result to form and output the historical total deviation value with individual identifiers.

[0008] Furthermore, the steps of determining the dynamic boundary based on historical parameters, comparing the current parameters with the dynamic boundary to obtain a comprehensive risk score, dividing the time window to extract feature vectors to calculate the trajectory deviation, and combining the parameter change rate to identify abnormal transitions include: Historical parameter data is read and multi-level quantiles are calculated according to time period. A dynamic boundary matrix is ​​constructed, and the current parameters are compared with the dynamic boundary to obtain the comprehensive risk score. The time window is divided based on the charge-discharge cycle. A window sequence is generated by sliding sampling. The statistical features of parameters within the window are extracted and combined. After standardization, a multidimensional feature vector is obtained. Construct a historical normal window baseline feature set, calculate the multi-class distance between the current feature vector and the baseline set, obtain the trajectory deviation after averaging and normalization, and mark abnormal windows based on dynamic thresholds; The parameter change rate is calculated point by point and the operation level is divided. The jumps and abnormal oscillations between levels are detected. The abnormal transition judgment is completed by combining the trajectory deviation and the relevant transition information is recorded.

[0009] Furthermore, the steps of dividing the time window based on the charge-discharge cycle, generating a window sequence using sliding sampling, extracting and combining the statistical features of the parameters within the window, and obtaining a multi-dimensional feature vector through standardization include: A single operating condition time window is divided by the charge and discharge cycle. A window sequence is generated by sliding sampling. Time series data of core parameters within the window are collected, and valid data is filtered and abnormal invalid data is removed. Calculate basic statistics for parameters within the window, perform frequency domain transformation on the basic statistics, generate corresponding derived statistics, and form a complete set of parameter features; The basic and derived statistics are combined hierarchically according to fixed dimensions to construct the original feature vector. The original feature vector is then subjected to dimensional unification processing to unify the vector format. The original feature vector is standardized according to dimensional statistics, the standardized values ​​are symmetrically truncated, and protection processing is completed by combining hierarchical labels, outputting a standardized multidimensional feature vector with window labels.

[0010] Furthermore, the steps of calculating the population deviation index using voltage, temperature, and SOC as dimensions to identify outlier subsets, and statistically analyzing the capacity decay rate and charge / discharge efficiency to obtain the capacity efficiency decay index, include: The distance between individual units is calculated using voltage, temperature, and SOC as dimensions. The group deviation index is determined based on the distance, and outlier subsets are identified based on the index. Collect the rated capacity and charge / discharge cycle data of individual cells, remove invalid samples, calculate the adjacent capacity decay rate and perform sign and constraint processing, and sum them to obtain the cumulative capacity decay rate. Calculate the total cumulative capacity decay and average decay level of individual cells, calculate the single charge-discharge efficiency and make interval correction, and determine the steady-state charge-discharge efficiency based on effective cycle data. Historical data on capacity decay rate and charge / discharge efficiency are extracted and relevant statistics are calculated. After standardization and nonlinear fusion, and then exponential transformation and percentage processing, the capacity efficiency decay index is obtained and output.

[0011] Furthermore, the steps of collecting the rated capacity and charge / discharge cycle data of individual cells, removing invalid samples, calculating adjacent capacity decay rates and performing sign and constraint processing, and summing them to obtain the cumulative capacity decay rate include: Collect charge-discharge cycle data of individual battery cells throughout their entire life cycle, verify the integrity of the data and the rationality of the operating conditions, remove invalid cycle samples, select valid cycles and organize them in an orderly manner to construct a continuous capacity dataset. Using adjacent effective cycles as the calculation unit and the capacity of the preceding cycle as the benchmark, a standardized adjacent capacity decay rate is obtained, and the decay rate is corrected for the aging stage. The signs of the corrected adjacent capacity decay rates are processed, the negative decay data corresponding to irreversible aging is retained, the positive fluctuation data is set to zero, and then the corresponding weights are assigned according to time and accumulated in a weighted manner. The single-cycle decay rate after successive accumulation is processed, and the monotonicity and boundary constraints of the cumulative result are applied. After normalization, the cumulative capacity decay rate with aging indicator is output.

[0012] Further, the step of obtaining a maintenance urgency index by weighted summation of the historical deviation total, comprehensive risk score, trajectory deviation degree, abnormal jumps, and outlier subset proportions, and outputting a graded maintenance decision based on the maintenance urgency index and capacity efficiency decay index, includes: Obtain relevant evaluation indicators such as historical total deviation and comprehensive risk score, classify and encode each indicator, combine them to form maintenance urgency code, and set graded judgment boundaries in combination with equipment operation attributes; Obtain capacity efficiency degradation indicators, classify and encode them, integrate the two types of codes to generate combined decision codes, conduct validity verification of relevant indicators, and screen out abnormal values; The combined decision code is decoded in a hierarchical manner, and the risk level is maintained according to the coding level. Boundary correction and verification are completed by combining the attenuation index. The decoded maintenance decisions are smoothed, matched with preset standardized maintenance plans, the maintenance operation requirements are clarified, and the hierarchical maintenance decision content is output.

[0013] The second aspect of this application also proposes a battery cabinet maintenance system based on fault prediction and health management, comprising: The multi-dimensional parameter processing module is used to collect multi-dimensional operating parameters of the battery cabinet and store them according to the operating stage, calculate the parameter variation coefficient and dynamic weight, and obtain the stage health index by weighted summation of parameter deviation. The individual health assessment module is used to obtain an individual health score based on the fusion of the phased health index, calculate and correct the health score by combining the historical out-of-limit frequency, and accumulate the deviation difference between the individual and the group median to obtain the historical deviation total value. The abnormal state identification module is used to determine the dynamic boundary based on historical parameters, compare the current parameters with the dynamic boundary to obtain a comprehensive risk score, divide the time window to extract feature vectors to calculate the trajectory deviation, and identify abnormal transitions by combining the parameter change rate. The group deviation analysis module is used to calculate the group deviation index in terms of voltage, temperature and SOC to identify outlier subsets, and to obtain the capacity efficiency degradation index by statistically analyzing the capacity decay rate and charge and discharge efficiency. The maintenance decision output module is used to obtain the maintenance urgency index by weighted summation of the total historical deviation value, comprehensive risk score, trajectory deviation degree, abnormal jump and outlier subset ratio, and output hierarchical maintenance decisions based on the maintenance urgency index and capacity efficiency decay index.

[0014] The first aspect of this plan brings the following benefits: This application comprehensively depicts the true health status of the battery cabinet by collecting multi-dimensional operating parameters, calculating health indices in stages, and assessing the health of individual cells. It also combines dynamic boundary recognition, trajectory deviation detection, and group deviation analysis to achieve early fault prediction and accurate anomaly location. Furthermore, it forms a hierarchical maintenance decision based on the maintenance urgency index and capacity efficiency decay index, transforming passive emergency repairs into proactive prevention. This effectively solves the problems of missed status assessments and delayed maintenance, inhibits accelerated cell aging, and improves the precision of battery cabinet operation and maintenance and operational safety. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a battery cabinet maintenance method based on fault prediction and health management according to an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a battery cabinet maintenance system based on fault prediction and health management according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer device according to an embodiment of this application; The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0017] Those skilled in the art will understand that, unless explicitly stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of this application means the presence of features, integers, steps, operations, elements, modules, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, modules, components, and / or groups thereof. It should be understood that when an element is “connected” or “coupled” to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein may include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any modules and all combinations of one or more associated listed items.

[0018] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0019] Reference Figure 1 This application provides a battery cabinet maintenance method based on fault prediction and health management, including: S1: Collect multi-dimensional operating parameters of the battery cabinet and store them according to the operating stage. Calculate the parameter variation coefficient and dynamic weight, and combine the parameter deviation with weighted summation to obtain the staged health index. S2: Based on the fusion of the phased health index, the individual health score is obtained, and the corrected health score is calculated by combining the historical over-limit frequency. The total historical deviation value is obtained by accumulating the deviation difference between the individual and the group median. S3: Determine the dynamic boundary based on historical parameters, compare the current parameters with the dynamic boundary to obtain a comprehensive risk score, divide the time window to extract feature vectors to calculate the trajectory deviation, and identify abnormal transitions by combining the parameter change rate. S4: Calculate the population deviation index using voltage, temperature and SOC as dimensions to identify outlier subsets, and obtain the capacity efficiency degradation index by statistically analyzing the capacity decay rate and charge / discharge efficiency. S5: The maintenance urgency index is obtained by weighted summation of the historical deviation total, comprehensive risk score, trajectory deviation degree, abnormal jump and outlier subset ratio. Based on the maintenance urgency index and capacity efficiency decay index, a graded maintenance decision is output.

[0020] In step S1, the reference object is first determined to be the single cell numbered #1 in the battery cabinet, and the current operating stage is "charging stage". Then, the input parameters for calculating the phased health index are prepared: the collected multi-dimensional operating parameters include voltage, temperature, current, and SOC (State of Charge), with a sampling period of 1 second and a current time window of 600 seconds (10 minutes); historical data includes the operating parameters of all "charging stages" of this cell over the past 90 days; the real-time parameters at the current moment (300th second) are: voltage 3.65V, temperature 32℃, current 100A, and SOC 75%. Then, the calculation is performed. First, the collected 600 seconds of data are categorized and stored according to the "charging stage" label. Then, the coefficient of variation (CV) of each parameter in the historical data is calculated: voltage CV is 0.02, temperature CV is 0.05, current CV is 0.1, and SOC CV is 0.03. The CV is normalized to obtain the dynamic weights wi: voltage weight 0.2, temperature weight 0.5, current weight 0.1, and SOC weight 0.2. Then, the deviation of the current parameters di is calculated: the voltage deviation is (3.65-3.60) / 3.60≈0.0139, the temperature deviation is (32-30) / 30≈0.0667, and the deviations of current and SOC are both 0. Finally, the weighted sum is used to obtain the staged health index HI: HI=(0.2×0.0139)+(0.5×0.0667)+(0.1×0)+(0.2×0)≈0.0361. This health index HI value quantifies the degree to which the overall health status of cell #1 deviates from the normal level in the current charging stage. The HI value of 0.0361 indicates that the cell has a slight abnormality in the current stage. Step S1 achieves a preliminary and accurate assessment of the health status of individual battery cells by quantifying, quantifying, and weighting the multidimensional parameters, providing calculable basic health indicators for subsequent steps.

[0021] In step S2, the calculation object is first determined to be the single cell of battery cabinet number #1. The health index of 0.0361 for each stage of the charging phase is used in step S1, and the charging phase health index remains unchanged during the operation phase. Then, the calculation input parameters are prepared: the health indices for single cells #1–#6 during the charging phase are 0.0361, 0.0122, 0.0095, 0.0153, 0.0217, and 0.0189 respectively; the voltage over-limit duration of cell #1 in the past 30 days is 28 minutes. The temperature exceeded the limit for 12 minutes, the current exceeded the limit for 0 minutes, and the total monitoring time was 43,200 minutes. The corrected health score set for the entire cabinet and individual cells was [85.2, 92.5, 93.8, 91.3, 88.7, 90.1]. Then, the calculation process was performed, and the health indexes at each stage were numerically transformed and weighted to obtain the health score of cell #1, S=86.3. The voltage over-limit frequency was calculated as 28 / 43200≈0.000648, and the temperature over-limit frequency was calculated as 43,200. Given a frequency limit of 12 / 43200 ≈ 0.000278 and a current over-limit frequency of 0, the weighted average over-limit frequency F ≈ 0.000371 is obtained by weighting by 0.4, 0.4, and 0.2. The corrected health score R is calculated as R = 86.3 × 0.000371 / 100 ≈ 0.000320. The median of the corrected health scores for the entire cabinet is taken, which is 90.7. The absolute value of the difference between cell #1 and the median is calculated as |0.000320 − 0.000358| = 0.000038, accumulated according to time decay weight, yields the total historical deviation value D=2.17 for cell #1. This value means: a single-cell health score of 86.3 indicates that cell #1's health level is low; a comprehensive over-limit frequency of 0.000371 indicates that the cell has a relatively low number of short-term over-limit occurrences; a corrected health score of 0.000320 is the comprehensive score after integrating health and over-limit values; and a total historical deviation value of 2.17 indicates the degree to which the cell has deviated from the normal state of the entire cabinet group over a long period. Step S2, through phased health index fusion, over-limit correction, comparison with the group median, and accumulation of historical deviations, transforms the individual cell's health status into a quantifiable score and cumulative deviation value, providing standardized individual status indicators for anomaly identification in step S3.

[0022] In step S3, the analysis object is first determined to be the battery cabinet #1 single cell. The operating parameters of the charging stage in step S1 and the calculation results in step S2 are used. The current window is a fixed 15-minute window in the charging stage. Then, the input parameters for calculation are prepared: the voltage dynamic boundary P10=3.52V, P90=3.72V and the temperature P10=28℃, P90=35℃ are obtained by statistically analyzing the historical parameters of 90 days by hour. The current voltage is 3.65V and the temperature is 32℃. The feature vector extracted from the window includes 12 dimensions: voltage mean, voltage change rate, temperature mean, and temperature rise rate. The benchmark feature vector library of the most recent 20 normal windows has been established. The parameter change rate thresholds L1=0.01 and L2=0.05. Then, the calculation process is executed, and the current parameters are compared with the dynamic boundary. The voltage and temperature are both in the normal range, and the comprehensive risk score R_total=0.23. Extract the 12-dimensional feature vector of the current window, calculate the average of Euclidean distance, cosine distance, and Manhattan distance with the benchmark library, and normalize to obtain a trajectory deviation DD=0.37; take the 95th percentile of the most recent 100 DDs as the dynamic threshold TH=0.32. If DD>TH, mark the current window as an abnormal window. Calculate the first-order change rate of voltage and temperature point by point. A voltage rate of 0.02, between L1 and L2, indicates a fluctuating layer; a temperature rate of 0.06, greater than L2, indicates a sudden change layer. If a temperature jump directly from a stable layer to a sudden change layer is detected, a judgment is made. An abnormal transition is defined as a transition path length L=1. The above values ​​mean: a comprehensive risk score of 0.23 indicates that the current parameter deviates from the normal range to a low degree; a trajectory deviation of 0.37 is greater than the threshold of 0.32, indicating that the running trajectory deviates significantly from the normal pattern; an abnormal transition path length of 1 indicates that there is a direct hierarchical jump anomaly. Step S3 constructs a dynamic boundary using 90 days of historical data, extracts features by dividing the window according to the charge and discharge cycle, calculates the trajectory deviation and identifies the rate transition, and converts the individual running status into a quantifiable risk value and anomaly marker.

[0023] In step S4, the calculation object is first determined to be the battery cell #1 in the battery cabinet. Using the charging stage data and anomaly judgment results from steps S1-S3, the sample data for all battery cells #1–#6 in the entire cabinet remains unchanged. Then, the calculation input parameters are prepared: the three-dimensional characteristic data of voltage, temperature, and SOC for each cell are as follows: #1 (3.65, 32, 75), #2 (3.61, 30, 74), #3 (3.60, 29, 73), #4 (3.62, 30, 74), #5 (3.63, 31, 74), #6 (3.62, 30, 74); the discharge capacity of cell #1 in the most recent 20 cycles is as follows: 49.8, 49.5, 49.2, 48.9, 48.7, 48.5, 48.3, 48.1, 47.9, 47.7, 47... The values ​​are 0.5, 47.3, 47.1, 46.9, 46.7, 46.5, 46.3, 46.1, 45.9, and 45.7. The corresponding total charging amount is 980.5, and the total discharging amount is 936.2. Then, the calculation process is performed to calculate the Euclidean distance between individual cells using three-dimensional features. The deviation index PDI of cell #1 is 1.82, the average PDI of the whole cabinet is 0.87, and the standard deviation is 0.52. Since 1.82 > 0.87 + 1.5 × 0.52, cell #1 is classified into the outlier subset. The proportion of the outlier subset is P = 1 / 6 ≈ 0.167. The average capacity decay rate of cell #1 is calculated as (45.7 - 49.8) / 49.8 / 19 ≈ -0.435%, and the average charge and discharge efficiency is 936.2 / 980.5 × 100% ≈ 95.5%. The capacity decay rate and charge / discharge efficiency are standardized and nonlinearly fused. After exponential transformation and percentage processing, the capacity efficiency decay index of cell #1 is obtained as 31 points. The above values ​​mean: the group deviation index of 1.82 indicates that cell #1 has poor consistency with the whole cabinet group; the proportion of outliers is 0.167, indicating that the individual cell has obvious outlier characteristics; the average capacity decay rate of -0.435% indicates that the cell capacity is showing a continuous downward trend; the average charge / discharge efficiency of 95.5% indicates that the energy conversion efficiency is at a normal to low level; and the capacity efficiency decay index of 31 points indicates that the cell aging has reached a moderate level. Step S4 identifies outlier cells through three-dimensional group deviation calculation, and forms a capacity efficiency decay index based on the cyclic data statistics of capacity decay and charge / discharge efficiency, quantifying the long-term aging of individual cells and the state of group deviation.

[0024] In step S5, the decision-making object is first determined to be the battery cabinet #1 single cell. Using all the calculation results from steps S1 to S4, the final maintenance decision calculation is completed using data from the same charging stage. Next, the input parameters are prepared: historical total deviation D = 2.17, comprehensive risk score R_total = 0.23, trajectory deviation DD = 0.37, abnormal transition path length L = 1, outlier subset percentage P = 0.167; time decay weights are 0.35, 0.25, 0.20, 0.15, and 0.0 respectively. 5; Capacity efficiency degradation index score: 31 points, average capacity degradation rate: -0.435%, average charge / discharge efficiency: 95.5%; Maintenance urgency index grading thresholds are: U<0.3 corresponds to M0, 0.3≤U<0.7 corresponds to M1, U≥0.7 corresponds to M2. Then, the calculation process is performed, and the five indicators are weighted and summed according to the time degradation weight to obtain the maintenance urgency index U=2.17×0.35+0.23×0.25+0.37×0.20+1×0.15+0.167×0.05=0.7595. The indicators are coded in a hierarchical manner. The historical total deviation, trajectory deviation, and abnormal transition path length are coded as 2. The comprehensive risk score and outlier subset ratio are coded as 0. The maintenance urgency is coded as 20220. The capacity efficiency decay index of 31 points is coded as 1. The combined decision is coded as 202201. The number of 2s in the statistical code is 3, which satisfies the condition 0.3≤U<0.7. Combining the capacity decay and efficiency values, the M1 planned maintenance decision is output. At the same time, cell #1 is marked as an outlier. The above values ​​mean: the maintenance urgency index of 0.7595 indicates that the maintenance level of cell #1 is too high; the combined decision code 202201 indicates that there are multiple medium-to-high risk abnormal features; the M1 planned maintenance indicates that the cell needs to be scheduled for regular maintenance and does not need to be replaced immediately. Step S5 calculates the maintenance urgency index by weighting the five indicators and completes the coding and hierarchical judgment by combining the capacity efficiency decay index. All the quantitative status data in the early stage are transformed into executable maintenance decisions, and the clear maintenance level and abnormal cell number are output.

[0025] In one embodiment, the steps of collecting multidimensional operating parameters of the battery cabinet and storing them according to operating stages, calculating the parameter variation coefficient and dynamic weight, and obtaining a staged health index by weighted summation of parameter deviations include: S10: Collect multi-dimensional operating parameters of the battery cabinet, divide the time sequence into different operating stages and form a stage parameter matrix, fill in missing data, and complete parameter classification and storage. S11: Calculate the coefficient of variation for the parameter matrix of each stage, introduce the time decay factor to obtain the weighted coefficient of variation, extract the comprehensive coefficient of variation and normalize it, and determine the dynamic weight after secondary correction and weight compensation. S12: Calculate the parameter deviation based on the stage statistical benchmark, perform truncation and fusion processing on the deviation, combine the parameter change information to obtain the final deviation, and feed back the abnormal frequency to the weight iteration. S13: Multiply the dynamic weights by the final absolute value of the deviation and sum them to obtain the phased health index for the current stage. Store the health indexes by stage and establish a time series sequence. Output the health index values ​​with stage labels.

[0026] In this embodiment, multi-dimensional operating parameters of the battery cabinet are first collected and stored according to the operating stage to form a stage parameter matrix. The collection objects are six individual battery cells (#1 to #6) in the battery cabinet. The collected parameters include individual cell voltage, individual cell temperature, charging and discharging current, SOC, cycle count, and internal resistance estimate. The sampling period is 1 minute. The current operating stage is the charging stage. Data from 30 consecutive minutes is extracted to form a charging stage parameter matrix. The matrix has 30 rows and 6 columns × 6 parameters = 36 columns. There are two instances of missing data at a single time point in the matrix, which are filled by the average of the effective values ​​of adjacent time points. There are no more than three consecutive missing time points, so linear interpolation is not required. After filling, the data is stored according to the charging stage label to obtain a clean and well-organized charging stage parameter matrix, which provides a unified data source for subsequent weight calculation. Next, the weighted coefficient of variation of the stage parameter matrix is ​​calculated and the dynamic weights are determined. The standard parameter sequence of the charging stage is read, and the basic coefficients of variation of voltage, temperature, current, SOC, internal resistance, and number of cycles are calculated respectively. A time decay factor is introduced, and the weight is higher the closer to the current moment. The weighted coefficients of variation of each parameter are obtained. The parameter co-variance matrix is ​​extracted to obtain the comprehensive variation index. After normalization, the initial weights are obtained. The frequency of each parameter falling outside the 95% confidence interval in the past 90 days is counted. The voltage has the highest frequency of exceeding the limit, so weight compensation is performed. The other parameters are corrected twice according to the excess ratio. All weights are normalized and constrained to a minimum value of 0.05. Finally, the dynamic weights of this stage are obtained: voltage 0.25, temperature 0.3, current 0.15, SOC 0.1, internal resistance 0.1, and number of cycles 0.1, with a total weight of 1.

[0027] The parameter deviations were calculated again based on the stage statistical benchmark and then truncated and fused. The historical average values ​​of the charging stage over 30 minutes were read: average voltage 3.60V, average temperature 30℃, average current 95A, average SOC 73%, average internal resistance 1.2mΩ, and average number of cycles 150. The deviations of each parameter of cell #1 at the current moment were calculated as follows: voltage (3.65-3.60) / 3.60≈0.0139, temperature (32-30) / 30≈0.0667, current (100-95) / 95≈0.0526, SOC (75-73) / 73≈0.0274, internal resistance (1.25-1.2) / 1.2≈0.0417, and the deviation of the number of cycles was 0. Since the absolute values ​​of all deviations were less than 3, no truncation was required, and they were directly used as the final deviations. The frequency of abnormalities was recorded simultaneously for subsequent weight iterations. Then, the dynamic weights are multiplied by the final deviation and summed to obtain the phased health index. The dynamic weights determined for this phase are multiplied by the final deviation of cell #1 item by item: Voltage 0.25×0.0139≈0.0035, Temperature 0.3×0.0667≈0.0200, Current 0.15×0.0526≈0.0079, SOC 0.1×0.0274≈0.0027, Internal Resistance 0.1×0.0417≈0.0042, Cycle Count 0.1×0=0. The six results are summed: 0.0035+0.0200+0.0079+0.0027+0.0042=0.0383, resulting in a phased health index of 0.0383 for cell #1 at the current charging stage. Finally, the health index is stored in stages and a time sequence is established. The index is marked with the charging stage, timestamp, and cell number #1 and stored in the health index time sequence queue. It is stored separately from the discharge, rest, and equalization stage indices. This index provides direct input for subsequent individual cell health scoring. The value objectively reflects the degree to which the cell parameters deviate from the normal level in the current stage. The higher the value, the higher the degree of abnormality. In this embodiment, multi-dimensional operating parameters are processed in layers, and the weights are dynamically matched to quantify parameter deviations. The health index is accurately generated in stages, and the operating status of the battery cabinet under various conditions is comprehensively and objectively characterized.

[0028] In one embodiment, the steps of calculating the coefficient of variation for the parameter matrix at each stage, introducing a time decay factor to obtain a weighted coefficient of variation, extracting and normalizing the comprehensive coefficient of variation, and determining the dynamic weights after secondary correction and weight compensation include: S110: Read the parameter matrix of each stage and sort it by time, set the corresponding time window, filter the valid data segment and remove abnormal data, and organize it into a standard parameter sequence according to the timestamp. S111: Calculate the basic coefficient of variation for the standard parameter sequence, assign time decay coefficients and calculate weighted statistics to obtain the weighted coefficient of variation, and extract the parameter covariance matrix to obtain the comprehensive variation index; S112: Statistical parameter abnormality correlation characteristics generate comprehensive correction coefficients, and combined with the degree of variation index, perform double correction on the weighted coefficient of variation, and apply weight lower limit constraints to obtain intermediate weight values; S113: Normalize the intermediate weight values, compensate for abnormally frequent parameters, normalize the compensated weights again, output the final dynamic weights, and store the weight iteration records.

[0029] In this embodiment, the charging stage parameter matrix is ​​first read and sorted in ascending order by timestamp. The charging stage time window is set to 30 minutes. The most recent 30 minutes of valid data segments are extracted. The sequence is traversed to remove outliers such as voltage exceeding 2.8V–4.2V, temperature exceeding -20℃–60℃, and current exceeding -100A–100A. There are no extreme outliers in this case. After rearranging by timestamp, a standard parameter sequence of 6 columns and 30 rows is formed, which is used for subsequent coefficient of variation calculation. Then, the basic coefficient of variation is calculated for the standard parameter sequence, time decay coefficients are assigned and weighted statistics are calculated. The mean and standard deviation are calculated for each column to obtain the basic coefficient of variation: voltage 0.02, temperature 0.05, current 0.10, SOC 0.03, internal resistance 0.04, and number of cycles 0.01. The attenuation coefficients are assigned in reverse chronological order, with the coefficient for the most recent 10 minutes being 1.0, the coefficient for the middle 10 minutes being 0.9, and the coefficient for the earliest 10 minutes being 0.8. The weighted mean and weighted standard deviation of each column are calculated using the coefficient weighting to obtain the weighted coefficient of variation: voltage 0.022, temperature 0.055, current 0.108, SOC 0.033, internal resistance 0.044, and cycle number 0.011. The covariance matrix of parameters is extracted according to the correlation between columns, and the comprehensive variation index is obtained as 0.075.

[0030] The abnormal parameter correlation characteristics were statistically analyzed again to generate a comprehensive correction coefficient. Double correction was performed on the weighted coefficient of variation and a weight lower limit was applied. The number of times the limit was exceeded in the past 90 days was counted: voltage 6 times, temperature 9 times, current 12 times, SOC 4 times, internal resistance 7 times, and cycle count 1 time. The comprehensive correction coefficients were generated as follows: voltage 1.0, temperature 1.2, current 1.5, SOC 0.9, internal resistance 1.1, and cycle count 0.8. The weighted coefficient of variation was multiplied by the corresponding correction coefficient to complete the double correction. A weight lower limit constraint of 0.05 was applied to obtain the intermediate weight values: voltage 0.055, temperature 0.066, current 0.162, SOC 0.050, internal resistance 0.048, and cycle count 0.008. The cycle count was automatically increased to 0.05. Then, the intermediate weight values ​​are normalized. For parameters with abnormally frequent errors, weight compensation is applied and normalization is performed again. The sum of the intermediate weights is 0.431. After normalization: voltage 0.127, temperature 0.153, current 0.376, SOC 0.116, internal resistance 0.111, cycle count 0.116. Current exceeds the limit most frequently, so compensation is increased by 0.05; temperature exceeds the limit second most frequently, so compensation is increased by 0.03; voltage increases compensation by 0.02, and the rest remain unchanged. The weights after compensation are: voltage... The following parameters were used: voltage 0.147, temperature 0.183, current 0.426, SOC 0.116, internal resistance 0.111, and cycle count 0.116. These were then normalized to obtain the final dynamic weights: voltage 0.15, temperature 0.19, current 0.43, SOC 0.11, internal resistance 0.10, and cycle count 0.02. The final dynamic weights were then output, and the iteration record was stored. The weights, correction coefficients, compensation values, and calculation times were stored in the iteration log according to the charging stage. This set of weights is directly used for subsequent deviation weighting calculations. The values ​​are fixed and traceable, providing a stable and unified weighting basis for the phased health index and ensuring the reproducibility of health assessment results.

[0031] In one embodiment, the step of obtaining an individual health score based on the phased health index fusion, calculating a corrected health score by combining historical exceedance frequencies, and accumulating the deviation difference between the individual and the population median to obtain the total historical deviation value includes: S20: Read the health index of each stage of operation and perform the corresponding numerical transformation. Assign weights according to the stage of operation and perform weighted fusion. After removing abnormal extreme values, obtain the individual health score. S21: Statistically analyze the historical parameters of individual units that exceed the limits, calculate the frequency of each item exceeding the limits by hierarchical weighting and introducing a time decay factor, and obtain the comprehensive frequency of exceeding the limits by weighted fusion. Then, calculate the corrected health score by combining it with the individual health score. S22: Extract the population modified health score and sort it to determine the population median. Calculate the absolute difference between the individual and the median. Apply an adjustment coefficient based on the historical deviation level to obtain the adjusted deviation difference. S23: Add the adjusted deviation difference to the historical cumulative deviation value, assign weights according to time proximity and calculate cumulatively, perform constraint processing on the cumulative result, and form and output the historical total deviation value with individual identifier.

[0032] In this embodiment, the health indices of cell #1 for the four stages of charging, discharging, resting, and balancing are first read, which are 0.0383, 0.0215, 0.0172, and 0.0124 respectively. Corresponding numerical transformations are performed on each index: the charging stage is logarithmically compressed to 0.021, the discharging stage retains its original value of 0.0215, the resting stage is square-rooted to 0.131, and the balancing stage is linearly mapped to 0.037. Weights are assigned according to the importance of each stage: charging 0.4, discharging 0.3, resting 0.2, and balancing 0.1. The weighted sum is 0.021×0.4+0.0215×0.3+0.131×0.2+0.037×0.1=0.0426. After removing extreme values ​​and mapping to a score of 0-100, the health score of cell #1 is 86.3 points, which is used to characterize the basic health level of the cell itself. Secondly, the historical over-limit information of cell #1 over the past 30 days was statistically analyzed. Voltage exceeded the limit for 28 minutes, temperature exceeded the limit for 12 minutes, and current did not exceed the limit. The total monitoring time was 43,200 minutes. The over-limit frequencies of each item were calculated: voltage 0.000648, temperature 0.000278, and current 0. They were weighted according to voltage 0.4, temperature 0.4, and current 0.2. After introducing a time decay factor, the comprehensive over-limit frequency was obtained as 0.000371. The individual cell health score of 86.3 was multiplied by the comprehensive over-limit frequency and then divided by 100 to obtain the corrected health score of 0.000320, which is used to reflect the true health level after the fusion of over-limit status.

[0033] The corrected health scores of individual cells #1 to #6 in the battery cabinet were extracted again, and were 0.000320, 0.000358, 0.000369, 0.000354, 0.000339, and 0.000348, respectively. The sequence was sorted from smallest to largest, and the average of the third and fourth scores was taken to obtain the population median of 0.000351. The absolute difference between cell #1 and the median was calculated as |0.000320-0.000351|=0.000031. Compared with the historical average deviation difference of 0.000025, the current difference is larger. Multiplying it by the adjustment coefficient of 1.5, the adjusted deviation difference of 0.000047 was obtained, which is used to reflect the degree of deviation of the individual cell relative to the population. Finally, the historical cumulative deviation value of cell #1, 2.123, was read. The deviation difference of 0.000047 from the current adjustment was added, and weights were assigned according to time: the most recent 10 values ​​had a weight of 1.0, the middle 20 values ​​had a weight of 0.8, and the earliest 20 values ​​had a weight of 0.6. After weighted summation, a value of 2.170 was obtained. An upper limit constraint was applied to the result, which was not greater than 1000, and three decimal places were retained to form the historical deviation total value of 2.170 with individual cell identification. This value is used to mark the degree of long-term deviation of an individual cell from the group. The larger the value, the more significant the deviation, providing a stable quantitative basis for subsequent anomaly identification and maintenance decisions.

[0034] In one embodiment, the steps of statistically analyzing historical parameter exceedance information of individual entities, calculating the exceedance frequency of individual items by hierarchical weighting and introducing a time decay factor, weighted fusion to obtain the comprehensive exceedance frequency, and calculating the corrected health score by combining it with the individual health score include: S210: Read the historical parameter exceedance records of a single entity, statistically analyze the related characteristics of exceedance, perform verification and nonlinear transformation on the single entity health score, and obtain the transformed health score; S211: Assign weighting coefficients according to parameter importance and degree of exceeding limits, calculate and correct the exceeding limit index using time decay rule, normalize and compress the exceeding frequency to obtain the relative exceeding frequency; S212: Integrate and correct the out-of-limit index and complete the secondary calibration to obtain the comprehensive out-of-limit frequency. Calculate the transformed health score and the relative out-of-limit frequency, and obtain the initial value of the corrected health score after attenuation correction and constraint. S213: Map the individual health score to the overall over-limit frequency, combine the historical score average with the previous time score for weighted smoothing, and output the corrected health score with timestamp after numerical constraints and precision processing.

[0035] In this embodiment, the historical parameter over-limit records of cell #1 over the past 30 days are first read. The statistics show that the voltage exceeded the limit for 28 minutes, the temperature exceeded the limit for 12 minutes, and the current exceeded the limit for 0 minutes, with a total monitoring time of 43,200 minutes. The health score of cell #1 is read as 86.3 points. The verification value is within the range of 0-100. Without boundary truncation, a nonlinear transformation is performed. 100 is subtracted from 86.3 to get 13.7. The square root is 3.70. Then 100 is subtracted from 3.70 to get the transformed health score of 96.30. This step completes the statistical analysis of over-limit features and the health score format transformation, providing a unified input for subsequent calculations. Next, weighting coefficients are assigned according to parameter importance: voltage 0.4, temperature 0.4, and current 0.2. Using a time decay rule, the weights are 1.0 for the last 7 days, 0.8 for 7–15 days, and 0.6 for 15–30 days. Corrected out-of-limit indicators are calculated as follows: voltage 0.000648, temperature 0.000278, and current 0. The three indicators are normalized by dividing by their respective historical maximum values ​​and then logarithmically compressed to obtain the relative out-of-limit frequencies: voltage 0.150, temperature 0.065, and current 0. This step completes the weighting and scaling of out-of-limit indicators, resulting in relative frequencies that can be directly fused.

[0036] The three corrected over-limit indicators are fused again, and the sum is obtained by weighting the coefficients to 0.000371. After secondary calibration, the comprehensive over-limit frequency of 0.000371 is determined. The transformed health score of 96.30 is weighted and multiplied by the relative over-limit frequency to obtain the initial corrected health score of 85.1. An upper limit constraint is imposed on the initial value, with the maximum value not exceeding 100 and the minimum value not lower than 0, so that the result remains within the valid range. This step completes the frequency fusion and initial value calculation, forming the unsmoothed base score. Then, the individual cell health score of 86.3 and the comprehensive over-limit frequency of 0.000371 are range-mapped to obtain the base value of 86.28. The average corrected health score of the past 5 times for cell #1 is read as 85.8. The current initial value of 85.1 is weighted with the historical average at a ratio of 0.6:0.4 to obtain 84.98. The score of the previous moment is read as 85.5. The current result is smoothed with the previous moment at a weighted ratio of 0.7:0.3 to obtain 85.04. The results are integerized, constrained to the range of 0–100, and the integer part is retained to obtain a score of 85. Finally, a current timestamp is added to this score, and a timestamped corrected health score of 85 is output. This score is simultaneously stored in the individual time series record table for subsequent calculation of the population median and statistics of historical deviations. This embodiment can objectively quantify the degree of exceeding the limit. Combined with time decay weighted smoothing, a stable and reliable corrected health score is output, improving the accuracy and consistency of individual state assessment.

[0037] In one embodiment, the steps of determining the dynamic boundary based on historical parameters, comparing the current parameters with the dynamic boundary to obtain a comprehensive risk score, dividing the time window to extract feature vectors to calculate the trajectory deviation, and combining the parameter change rate to identify abnormal transitions include: S30: Read historical parameter data and calculate multi-level quantiles according to time period, construct dynamic boundary matrix, compare current parameters with dynamic boundary, and obtain comprehensive risk score; S31: Divide the time window based on the charge and discharge cycle, generate a window sequence by sliding sampling, extract and combine the statistical features of parameters within the window, and obtain a multi-dimensional feature vector after standardization. S32: Construct a historical normal window baseline feature set, calculate the multi-class distance between the current feature vector and the baseline set, obtain the trajectory deviation after averaging and normalization, and mark abnormal windows according to the dynamic threshold; S33: Calculate the parameter change rate point by point and divide the operation level, detect cross-level jumps and abnormal oscillations between levels, combine the trajectory deviation to complete the abnormal transition judgment, and record the transition-related information.

[0038] In this embodiment, the first step is to read historical parameter data and calculate multi-level quantiles by time period to construct a dynamic boundary matrix. Historical charging data for cell #1 over the past 90 days is read, and the P10 and P90 quantiles of voltage and temperature are calculated in 24-hour segments to form a dynamic boundary matrix: voltage P10 = 3.52V, P90 = 3.72V, temperature P10 = 28℃, P90 = 35℃. The current parameters of voltage 3.65V and temperature 32℃ are compared with the corresponding time period boundaries, and both fall within the normal range. The number and magnitude of exceedances are counted to obtain a comprehensive risk score of 0.23, which is stored according to the cell number to provide a benchmark for subsequent trajectory deviation calculations. Secondly, the time window is divided based on the charge and discharge cycle. A sliding sampling method is used to generate a window sequence, extract features and standardize them. The charging phase is 15 minutes as a single working condition window, and a sliding step of 7.5 minutes is used to generate a window sequence. Voltage, temperature, current and SOC time series data are collected within the current window. Communication anomalies and physical over-limit points are removed. Statistics such as mean, range, rate of change and equivalent resistance are calculated and combined to form a 12-dimensional original feature vector. Each dimension is divided by the historical maximum value to unify the dimensions. Then the mean of the dimensions is subtracted and divided by the absolute median difference to obtain a standardized multi-dimensional feature vector, which is used for trajectory distance calculation.

[0039] Next, a baseline feature set of historical normal windows is constructed. Multi-class distances are calculated to obtain the trajectory deviation. Feature vectors from the 20 most recent normal windows during the charging phase are retrieved to form a baseline set. The Euclidean distance, cosine distance, and Manhattan distance between the current vector and the baseline are calculated, averaged, and normalized to obtain a trajectory deviation of 0.37. The most recent 100 trajectory deviations are sorted, and the 95th percentile is used as the dynamic threshold of 0.32. If the current 0.37 > 0.32, the current window is marked as an abnormal window, and the window number and timestamp are recorded. This process provides a basis for identifying abnormal transitions. Finally, the parameter change rate is calculated point-by-point to classify operational levels and detect abnormal transitions. The first-order change rates of voltage and temperature are calculated point-by-point, with L1=0.01 and L2=0.05. A voltage rate of 0.02 is classified as a fluctuation layer, and a temperature rate of 0.06 as a sudden change layer. Tracing back three moments, if the temperature transitions directly from the stable layer to the sudden change layer, it is identified as an abnormal transition. The transition path length is recorded as 1, and combined with the trajectory deviation anomaly marker, the validity of this transition is confirmed and stored in the cell event list. This step, through fixed quantile boundaries, sliding windows, multi-distance fusion, and rate-level judgment, forms a reproducible anomaly identification process. The output comprehensive risk score, trajectory deviation, anomaly window marker, and abnormal transition path are all integrated into the subsequent maintenance urgency index calculation, providing an objective quantitative basis for graded maintenance decisions.

[0040] In one embodiment, the steps of dividing the time window based on the charge-discharge cycle, generating a window sequence using sliding sampling, extracting and combining statistical features of parameters within the window, and obtaining a multi-dimensional feature vector through standardization include: S310: Divide the time window of a single operating condition by the charge and discharge cycle, use sliding sampling to generate a window sequence, collect the time series data of the core parameters within the window, filter valid data and remove abnormal and invalid data; S311: Calculate the basic statistics for the parameters within the window, perform frequency domain transformation on the basic statistics, generate corresponding derived statistics, and form a complete set of parameter features; S312: Combine the basic statistics and derived statistics in a fixed dimension to construct the original feature vector, and perform dimensional unification processing on the original feature vector to unify the vector format; S313: Perform standardization calculations on the original feature vector according to dimensional statistics, symmetrically truncate the standardized values, and complete the protection process by combining hierarchical labels, outputting a standardized multidimensional feature vector with window labels.

[0041] In this embodiment, the first step is to divide the time window of a single operating condition by the charge and discharge cycle. A sliding sampling method is used to generate a window sequence. The time series data of the core parameters are collected and cleaned. Taking the charging stage of cell #1 as the object, the single operating condition window is divided into 15-minute intervals. The sliding step size is half the window length, i.e., 7.5 minutes, to generate a continuous, non-overlapping, and partially covered window sequence. The four core parameters of voltage, temperature, current, and SOC are collected within the current window, one point per minute, for a total of 15 time series data points. Communication verification, physical rationality verification, and logical consistency verification are performed on the data in sequence. Abnormal and invalid points are removed, and missing values ​​are replaced to obtain a clean, continuous, and effective time series data segment, providing a reliable data source for feature extraction. Secondly, basic statistics are calculated for the effective parameters within the window, and frequency domain transformation is performed to generate derived statistics. The maximum, minimum, mean, and standard deviation of voltage, temperature, current, and SOC are calculated respectively. Based on this, frequency domain transformation is performed on the time-series signal to calculate the equivalent resistance by the ratio of voltage integral to current integral, the heat charge product by the product of the effective temperature value and SOC integral, the peak frequency by the ratio of the number of current peaks to the window duration, and the fundamental frequency by the ratio of the number of voltage zero crossings to the window duration. This forms a complete set of parameter features including basic statistics and frequency domain derived quantities, with a total of 12 feature items, ensuring that the feature dimensions are fixed and can be calculated repeatedly.

[0042] The next step involves combining basic and derived statistics in a fixed-dimensional hierarchical manner to construct an original feature vector with unified dimensions. Following a fixed order of voltage, temperature, current, and SOC, 12 types of features, including mean, equivalent resistance, peak frequency, heat charge product, and fundamental frequency, are arranged in a hierarchical manner to construct a 12-dimensional original feature vector. Dimensional unification is performed on each feature dimension by dividing the current value by the historical maximum value of that dimension, mapping all values ​​to the interval between 0 and 1. This ensures that the magnitudes of different dimensions and parameters are consistent, eliminates calculation biases caused by unit differences, and forms an original feature vector with a unified format, fixed length, and direct comparability. Finally, the original feature vectors are standardized, symmetrically truncated, and protected, outputting standardized multidimensional feature vectors with window labels. The global mean and absolute median are calculated dimension-by-dimensionally for the 12-dimensional original feature vectors. The standardization is completed by subtracting the corresponding mean from each dimension's value and then dividing by the absolute median. Symmetrical truncation is performed on the standardized results: values ​​greater than 2 are set to 2, and values ​​less than -2 are set to -2 to avoid interference from extreme values. A protection coefficient is assigned based on the current window level label, ultimately generating a stable, regular, and reproducible standardized multidimensional feature vector. This vector is then appended with the current window number, timestamp, and operating condition stage identifier and stored in the feature database for subsequent trajectory deviation calculation and abnormal window identification. This step, through objective processes such as fixed window division, sliding sampling, data cleaning, statistical calculation, frequency domain transformation, vector combination, unit unification, standardization, and truncation, outputs a fixed-dimensional, comparable, and reproducible standardized feature vector, providing a stable and unified quantitative feature foundation for battery cabinet status identification.

[0043] In one embodiment, the steps of dividing a single operating condition time window by charge / discharge cycle, generating a window sequence using sliding sampling, collecting time-series data of core parameters within the window, filtering valid data, and removing abnormal and invalid data include: S3101: Divide a single operating condition time window by combining the charge and discharge cycle with the operating condition switching node, use sliding sampling to generate a window sequence, and collect the original time series data of parameters within the window; S3102: Perform communication quality verification and sensor validity verification on the raw time series data, perform data inference and replacement on abnormal data points, and mark and remove data from abnormal time periods; S3103: Perform physical rationality verification and logical consistency verification on the verified data, replace abnormal data points, and filter and remove abnormal data with parameter jumps and timing breaks. S3104: Centrally remove invalid data and complete short-term missing data, perform final data integrity review and freshness verification, replace and downgrade abnormal data, and output clean window parameter time series data.

[0044] In this embodiment, the first step is to divide a single operating condition time window by combining the charge / discharge cycle with the operating condition switching node. A sliding sampling method is used to generate a window sequence and collect the original time-series data of the parameters within the window. Taking the charging stage of cell #1 as the analysis object and the complete charge / discharge cycle as the benchmark, the boundary is aligned at the operating condition switching node to divide a 15-minute single operating condition time window. A sliding sampling method is used with a step size of 7.5 minutes to generate a continuous and partially overlapping window sequence. The original time-series data of the four core parameters of voltage, temperature, current, and SOC within the current window are collected. The sampling interval is 1 minute, and a total of 15 data points are collected to form the original time-series dataset, which provides the original input for subsequent verification. Secondly, communication quality and sensor validity checks are performed on the original time series data. Abnormal data points are replaced by data inference. Abnormal time periods are marked and removed. The 15 original data points are traversed, the communication flag bits are checked, and two communication abnormal points are identified. Replacement is completed by linear interpolation of adjacent normal data points. The sensor status flags are checked to confirm that all sensors are in normal condition and there are no faulty channels. Time periods with more than 5 consecutive abnormal points are marked. There are no such time periods in this window, so there is no need to remove the entire segment. After verification and replacement, time series data with qualified communication quality and valid sensors are obtained, eliminating the calculation deviation caused by transmission anomalies.

[0045] Next, physical rationality and logical consistency checks were performed on the verified data. Abnormal data points were replaced, and abnormal data with parameter jumps and timing breaks were filtered and removed. The physical ranges were set as follows: voltage 2.8V–4.2V, temperature -20℃–60℃, current -100A–100A, and SOC 0–100%. One voltage point was found to exceed the upper limit, and it was replaced with the previous normal data point. Logical consistency was checked to confirm that the direction of SOC change and the direction of current were completely matched and there were no logical contradictions. The parameter jump amplitude was checked, and there were no jump points exceeding the gradient threshold and no timing breaks. After the verification was completed, all 15 data points were retained, and the data met the physical and logical constraints. Finally, invalid data is centrally removed and short-term missing data is supplemented. A final review of data integrity and freshness is performed. Abnormal data is replaced and its weight is downgraded. Clean window parameter time-series data is output. This window has no invalid data to remove, and all short-term missing points have been supplemented. The coefficient of variation for each parameter is calculated; there are no constant values ​​or drastic fluctuations. Data freshness is checked; the acquisition time difference for all points is less than 50% of the window duration, so no time-sensitivity downgrade is needed. After final review, clean, continuous, and reliable window parameter time-series data is formed. A window number, timestamp, and operating condition label are added to this data segment, and it is stored in the time-series database as the sole valid input for subsequent statistical calculations and feature extraction. This step, through a four-layer verification process with fixed rules, completes the cleaning, correction, supplementation, and final review of the original data, providing a reliable data source for subsequent basic statistics, frequency domain derived quantities, and feature vectors, ensuring that the battery cabinet status identification results are authentic, reproducible, and traceable.

[0046] In one embodiment, the steps of calculating the population deviation index using voltage, temperature, and SOC as dimensions to identify outlier subsets, and statistically analyzing the capacity decay rate and charge / discharge efficiency to obtain a capacity efficiency decay index, include: S40: Calculate the distance between individual units using voltage, temperature and SOC as dimensions, determine the group deviation index based on the distance, and screen and identify outlier subsets based on the index; S41: Collect the rated capacity and charge / discharge cycle data of individual cells, remove invalid samples, calculate the adjacent capacity decay rate and perform sign and constraint processing, and sum them to obtain the cumulative capacity decay rate. S42: Calculate the total cumulative capacity decay of a single cell and the average decay level, calculate the single charge-discharge efficiency and make interval corrections, and determine the steady-state charge-discharge efficiency based on effective cycle data. S43: Extract historical data on capacity decay rate and charge / discharge efficiency and calculate relevant statistics. After standardization and nonlinear fusion, and then exponential transformation and percentage processing, obtain and output the capacity efficiency decay index.

[0047] In this embodiment, the distance between individual cells is first calculated using voltage, temperature, and SOC as dimensions. Based on the distance, the group deviation index is determined and outlier subsets are selected. Taking six cells from battery cabinet #1 to #6 as the analysis object, the three-dimensional feature data at the current moment are taken as follows: #1(3.65,32,75), #2(3.61,30,74), #3(3.60,29,73), #4(3.62,30,74), #5(3.63,31,74), #6(3.62,30,74). The pairwise distances are calculated for each group. Using Euclidean distance, the average distance di_avg of cell #1 is 0.382, the global average distance d_global is 0.213, and the group deviation index PDI is 0.382 / 0.213=1.82. The calculated group PDI mean is 0.87 and the standard deviation is 0.52. The judgment threshold is 0.87+1.5×0.52=1.65. Cell #1's PDI=1.82>1.65, so it is classified into the outlier subset. The proportion of outliers in the entire cabinet is P=1 / 6≈0.167, which will be used for subsequent maintenance urgency index calculation. Secondly, the rated capacity and charge / discharge cycle data of individual cells are collected, invalid samples are removed, adjacent capacity decay rates are calculated and their signs and constraints are processed, and the cumulative capacity decay rate is obtained by summing them. The rated capacity of cell #1 is read as 50Ah, and the discharge capacity of the most recent 20 effective cycles is collected: 49.8, 49.5, 49.2, 48.9, 48.7, 48.5, 48.3, 48.1, 47.9, 47.7, 47.5, 47.3, 47.1, 46.9, 46.7, 46.5, 46.3, 46.1, 45.9, 45.7. After removing abnormal operating condition samples, the adjacent capacity decay rate is calculated one by one, only negative decay values ​​are retained, and positive fluctuations are set to 0. The cumulative capacity decay rate is obtained by weighting and summing according to the time decay weight, which is -0.082. The normalized average single decay rate is -0.435%, which is used to characterize the aging rate of the cell.

[0048] Next, the total cumulative capacity degradation and average degradation level of each individual cell were calculated. The single-cycle charge / discharge efficiency was calculated and adjusted for ranges to determine the steady-state charge / discharge efficiency. For cell #1, the total charge capacity after 20 cycles was 980.5 kWh, and the total discharge capacity was 936.2 kWh. The single-cycle efficiency was adjusted according to the SOC range: 0–20% coefficient 0.85, 20–80% coefficient 1, and 80–100% coefficient 0.9. After adjustment, the values ​​were summed to obtain a steady-state charge / discharge efficiency of 95.5%. Comparing the cumulative degradation with the rated capacity, the cumulative capacity degradation ratio was 8.2%, objectively reflecting the cell capacity loss. The degree of loss provides input for the capacity efficiency degradation index. Finally, historical data on capacity degradation rate and charge / discharge efficiency are extracted, and after standardization, nonlinear fusion, exponential transformation, and percentage processing, the capacity efficiency degradation index is output. The average capacity degradation rate of -0.435% and steady-state charge / discharge efficiency of 95.5% are standardized and mapped to a unified interval. The data is then fused using the formula: (|degradation rate|^0.5×0.7)+((100-efficiency) / 100×0.3)=0.055×0.7+0.045×0.3=0.0385+0.0135=0.052. After exponential transformation, 1−e^(-0.052)=0.0507, multiplied by 100 and rounded, resulting in a capacity efficiency degradation index of 31 points. This index is a percentage value and is directly used for the hierarchical maintenance decision in step S5, forming a complete input along with the historical deviation total value, comprehensive risk score, trajectory deviation degree, abnormal transition, and outlier percentage. This step, through fixed three-dimensional distance calculation, outlier identification, capacity decay statistics, efficiency correction, and indicator fusion, outputs quantifiable population deviation results and capacity efficiency decay indicators. This embodiment can accurately identify outlier individuals, objectively reflect capacity decay and efficiency changes, quantify the degree of health decay, and provide a stable and reliable basis for maintenance decisions.

[0049] In one embodiment, the steps of collecting the rated capacity and charge / discharge cycle data of individual cells, removing invalid samples, calculating adjacent capacity decay rates and performing sign and constraint processing, and summing them to obtain the cumulative capacity decay rate include: S410: Collect charge and discharge cycle data of individual battery cells throughout their entire life cycle, verify the integrity of the data and the rationality of the operating conditions, remove invalid cycle samples, select valid cycles and organize them in an orderly manner, and construct a continuous capacity dataset. S411: Using adjacent effective cycles as the calculation unit and the capacity of the preceding cycle as the benchmark, the standardized adjacent capacity decay rate is calculated, and the decay rate is corrected for the aging stage. S412: The signs of the corrected adjacent capacity decay rates are processed, the negative decay data corresponding to irreversible aging is retained, the positive fluctuation data is set to zero, and then the corresponding weights are assigned according to time and accumulated in a weighted manner. S413: Single-cycle decay rate after successive accumulation processing. The cumulative result is subjected to monotonicity and boundary constraints, and after normalization, the cumulative capacity decay rate with aging indicator is output.

[0050] In this embodiment, the first step is to collect the full lifecycle charge-discharge cycle data of individual battery cells, verify the data integrity and the rationality of the operating conditions, eliminate invalid cycle samples, select valid cycles and organize them in an orderly manner to construct a continuous capacity dataset. Taking cell #1 as the analysis object, with a rated capacity of 50Ah, the original records of the most recent 200 cycles are collected, and the integrity of the fields and the operating conditions are checked one by one. Invalid samples with incomplete charging, insufficient discharging, sensor malfunction, temperature exceeding the limit, and human intervention are eliminated. Finally, 20 sets of valid cycle data are selected and the discharge capacity sequence is arranged in chronological order: 49.8, 49.5, 49.2, 48.9, 48.7, 48.5, 48.3, 48.1, 47.9, 47.7, 47.5, 47.3, 47.1, 46.9, 46.7, 46.5, 46.3, 46.1, 45.9, 45.7, forming a time-continuous and complete standardized capacity dataset, providing reliable input for subsequent attenuation calculations. Secondly, using adjacent effective cycles as the calculation unit and the capacity of the preceding cycle as the benchmark, a standardized adjacent capacity decay rate is obtained. An aging stage correction is then applied to the decay rate. Using every two adjacent cycles as the unit, and based on the discharge capacity of the previous cycle, the adjacent decay rate is calculated as (later capacity – earlier capacity) / earlier capacity, resulting in the following values: -0.0060, -0.0061, -0.0061, -0.0061, -0.0041, -0.0041, -0.0041, -0.0041, -0.0 042, -0.0042, -0.0042, -0.0042, -0.0042, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, the current cycle count is 350, which is in the range of 200-500 cycles. The aging correction coefficient is 1.2. Multiply the above decay rates one by one by 1.2 to obtain the corrected adjacent capacity decay rate sequence, and complete the aging stage adaptation process.

[0051] Next, the signs of the adjacent capacity decay rates after correction are processed, retaining the negative decay data corresponding to irreversible aging and setting the positive fluctuation data to zero. Then, corresponding weights are assigned according to time and weighted accumulation is performed. The corrected sequence is traversed. All values ​​are negative decays with no capacity rebound fluctuations. All values ​​are retained in their original form and not set to zero. Weights are assigned according to time proximity. The weight of the most recent 10 times is 1.0, the weight of the middle 5 times is 0.8, and the weight of the earliest 4 times is 0.6. Each decay rate is multiplied by its corresponding weight to obtain a weighted single-cycle decay sequence, which is used to highlight the impact of recent decay on the current state, thus completing the weighted processing flow. Finally, the single-cycle decay rate after successive accumulation is calculated. Monotonicity and boundary constraints are applied to the accumulated result, and after normalization, the cumulative capacity decay rate with an aging indicator is output. The weighted single-cycle decay rates are then successively accumulated to obtain a cumulative decay result of -0.082. Monotonicity constraints are applied to ensure the cumulative value only decreases and does not increase; boundary constraints are applied to limit the value to the range [-0.2, 0]. No adjustment is needed for this result. The absolute cumulative value is divided by the effective number of cycles to obtain the normalized average single-cycle decay rate of -0.435%. The result is labeled with the aging stage indicator "mid-term aging," forming the cumulative capacity decay rate with an aging indicator. This step, through a fixed process of data filtering, decay calculation, aging correction, sign constraint, weighted accumulation, and boundary normalization, completely outputs the cumulative capacity decay rate, providing a stable and reliable quantitative basis for battery cabinet fault prediction and health management.

[0052] In one embodiment, the step of calculating a standardized adjacent capacity decay rate based on adjacent effective cycles and prior cycle capacity, and then performing aging stage correction on the decay rate, includes: S4110: Select two adjacent effective cycles as the calculation unit, take the capacity of the earlier cycle as the benchmark, quantify the capacity difference between the later cycle and the benchmark, and obtain the standardized single-cycle capacity decay rate through relative proportion conversion. S4111: Extract the capacity data of each cycle, organize the single-cycle decay data in chronological order, and form a continuous single-cycle decay sequence. S4112: Divide the aging stages according to the number of cycles, assign corresponding correction coefficients to the capacity decay rate of different stages, and perform aging stage adaptation correction on the single-cycle decay rate to form corrected adjacent decay data. S4113: Verify the validity of the corrected adjacent attenuation rates, remove outliers, sort them in chronological order to form a continuous and complete corrected attenuation sequence, and output the aging-corrected adjacent attenuation rates with cycle count identifier.

[0053] In this embodiment, two adjacent effective cycles are first selected as the calculation unit. Using the capacity of the earlier cycle as a benchmark, the capacity difference between the later cycle and the benchmark is quantified. Through relative percentage conversion, a standardized single-cycle capacity decay rate is obtained. Using the 20 groups of effective discharge capacities already screened for cell #1 as the calculation object, adjacent groups of data are taken sequentially. The capacity of the previous cycle is used as the benchmark value, and the capacity of the later cycle is used as the current value. The 19 single-cycle capacity decay rates are calculated sequentially using the formula (current capacity - benchmark capacity) / benchmark capacity: -0. .0060, -0.0061, -0.0061, -0.0061, -0.0041, -0.0041, -0.0041, -0.0042, -0.0042, -0.0042, -0.0042, -0.0042, -0.0042, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, -0.0043, all attenuation rates are in relative value format, with unified dimensions, forming a basic attenuation sequence that can be directly corrected. Secondly, the capacity data of each cycle is extracted, and the single-cycle decay data is organized in chronological order to form a continuous single-cycle decay sequence. The above 19 single-cycle capacity decay rates are arranged in the order in which the cycles occur, without changing the order, removing values, or performing smoothing. The original calculation results are directly retained. The sequence starts at the 331st cycle and ends at the 350th cycle, which is completely aligned with the time sequence of the original capacity data. After arrangement, a continuous single-cycle decay sequence of length 19 is formed, with each position corresponding to the cycle number one by one, without breaks, repetitions, or misalignments. This ensures that the decay sequence and the cycle history are consistent in time, providing an alignment basis for subsequent aging stage corrections.

[0054] Next, the aging stages are divided according to the number of cycles, and corresponding correction coefficients are assigned to the capacity decay rate of different stages. The single-cycle decay rate is adjusted for the aging stage to form the corrected adjacent decay data. The current cumulative number of cycles of cell #1 is 350. The aging stages are divided according to the rules: less than 200 cycles is the initial stage with a coefficient of 1.0; 200 to 500 cycles is the middle stage with a coefficient of 1.2; and more than 500 cycles is the late stage with a coefficient of 1.5. All cycles in this case are in the middle stage range of 200-500 cycles, and a correction coefficient of 1.2 is uniformly adopted. Multiply each value in the single-cycle decay sequence by 1.2 to obtain the corrected sequence: -0.0072, -0.0073, -0.0073, -0.0073, -0.0049, -0.0049, -0.0049, -0.0050, -0.0050, -0.0050, -0.0050, -0.0050, -0.0052, -0.0052, -0.0052, -0.0052, -0.0052, -0.0052, -0.0052. Finally, the validity of the corrected adjacent decay rates is verified, outliers are removed, and the values ​​are sorted in chronological order to form a continuous and complete corrected decay sequence. The aging-corrected adjacent decay rates are output with the cycle number identifier. The corrected sequence values ​​were checked one by one. All values ​​were within the range of [-0.01, 0], with no values ​​exceeding the upper limit, no reverse anomalies, and no sudden jumps. All values ​​were determined to be valid and did not need to be discarded. The values ​​were then rearranged and sorted according to the original time order, maintaining a one-to-one correspondence with the number of cycles. The starting cycle number 331 was marked at the beginning of the sequence, and the ending cycle number 350 was marked at the end, forming an adjacent decay sequence after aging correction with complete cycle number markings. This embodiment can standardize the calculation of capacity decay rate. Combined with aging stage correction, the values ​​are accurate and reliable, and can objectively reflect the aging state of the battery cell.

[0055] In one embodiment, the step of obtaining a maintenance urgency index by weighted summation of the historical deviation total, comprehensive risk score, trajectory deviation degree, abnormal jumps, and outlier subset proportion, and outputting a graded maintenance decision based on the maintenance urgency index and capacity efficiency decay index, includes: S50: Obtain relevant evaluation indicators such as historical total deviation and comprehensive risk score, perform hierarchical coding on each indicator, combine them to form maintenance urgency code, and set hierarchical judgment boundaries in combination with equipment operation attributes; S51: Obtain capacity efficiency decay indicators, classify and encode them, integrate the two types of codes to generate combined decision codes, perform validity verification on relevant indicators, and screen out abnormal values. S52: Perform hierarchical decoding of the combined decision code, divide and maintain the risk level according to the coding level, and complete the boundary correction and verification in combination with the attenuation index; S53: Smooth the decoded maintenance decisions, match them with preset standardized maintenance plans, clarify maintenance operation requirements, and output hierarchical maintenance decision content.

[0056] In this embodiment, the first step is to obtain five evaluation indicators: total historical deviation, comprehensive risk score, trajectory deviation degree, abnormal transition path length, and outlier subset percentage. Each indicator is then graded and coded to form a maintenance urgency code. Taking cell #1 as the calculation object, the values ​​of the five indicators are read: total historical deviation 2.170, comprehensive risk score 0.23, trajectory deviation degree 0.37, abnormal transition path length 1, and outlier subset percentage 0.167. These are coded according to the graded rules: 0 for low risk, 1 for medium risk, and 2 for high risk. A total historical deviation ≥ 1.5 is coded as 2; a comprehensive risk score < 0.3 is coded as 0; a trajectory deviation degree > 0.32 is coded as 2; an abnormal transition path length ≥ 1 is coded as 2; and an outlier subset percentage ≥ 0.1 is coded as 2. This combination yields a 5-digit maintenance urgency code of 20222. Based on the mid-term operational attributes of the battery cabinet, the graded judgment boundary is set with a fixed threshold and is not dynamically adjusted. The next step is to obtain the capacity efficiency degradation index, perform hierarchical coding on it, integrate the two types of codes to generate a combined decision code, and conduct validity verification. The capacity efficiency degradation index of #1 cell is 31 points, and it is hierarchically coded according to the rules: <10 is 0, 10-30 is 1, >30 is 2, and the current code is 1. The maintenance urgent code 20222 is integrated with the degradation code 1 to generate a 6-bit combined decision code 202221. The validity of all indicators is verified to check whether the values ​​are within a reasonable range and whether the timing is aligned. It is confirmed that there are no abnormal jumps, no out-of-range values, and no missing items. After filtering out invalid values, the valid codes are retained for subsequent hierarchical decoding.

[0057] Next, the combined decision code is decoded hierarchically. Risk levels are maintained based on the coding level, and boundary correction and verification are completed using attenuation indicators. The number of digits 2 in the 6-bit code is counted (4 in total). According to the decoding rules: <2 digits for M0, 2–4 digits for M1, and ≥4 digits for M2. Initially, it is determined to be at level M1. Considering the capacity efficiency attenuation indicator score of 31, with an average capacity attenuation rate of -0.435% and an average charge / discharge efficiency of 95.5%, it does not simultaneously meet the conditions of attenuation rate <-0.5% and efficiency <90%, thus not triggering the M2 boundary condition. Boundary correction is completed, and the M1 risk level is ultimately maintained. The matching of the coding rules and thresholds is verified, confirming that the decoding process conforms to a fixed procedure without subjective adjustments. Finally, the decoded maintenance decisions are smoothed, matched with a pre-defined standardized maintenance plan, clarifying maintenance requirements, outputting tiered maintenance decision content, checking the decision codes at adjacent times (no more than 3 bits change, no decision jumps, no additional smoothing required), matching the M1 level standardized plan: conduct planned maintenance, check the electrical connections and temperature measurement point contact status of outlier cells, verify voltage consistency and internal resistance data, output decision content: M1 planned maintenance, indicating that #1 cell is an outlier cell, marking the three main causes: historical deviation, trajectory deviation, and abnormal transition, and packaging and pushing the decision code, cell number, maintenance content, and timestamp to the operation and maintenance work order system. This step, through fixed indicator coding, combined decision, tiered decoding, boundary verification, and smoothed output process, objectively generates tiered maintenance conclusions, providing clear and actionable quantitative basis for preventive maintenance of battery cabinets.

[0058] In one embodiment, the steps of obtaining the capacity efficiency degradation index, hierarchically encoding it, integrating the two types of codes to generate a combined decision code, validating the relevant indexes, and filtering out outliers include: S510: Integrate time-series data of capacity efficiency decay index and maintenance urgency index, complete time-series alignment and regular arrangement, remove redundant and disordered data, identify and mark sequence anomalies through sliding window operation, and build a standardized basic dataset. S511: Conduct dual verification of time sequence structure and continuity, distinguish different types of abnormal data, use an adaptation method to complete the replacement and correction of abnormal values, and complete basic anomaly screening in combination with operating conditions. S512: Perform trend identification and autocorrelation analysis on the indicator sequence, screen for abnormal fluctuations and lagged anomalies, and complete deep anomaly identification and numerical correction based on neighborhood verification and comparison with contemporaneous data. S513: Purify the time series data of the two types of indicators and perform hierarchical coding respectively, remove abnormal disturbances in the coding sequence, match and combine the two sets of codes one by one according to the time sequence, and generate a standardized and unified combined decision coding sequence.

[0059] In this embodiment, the first step is to integrate the time-series data of capacity efficiency degradation index and maintenance urgency index, complete time-series alignment and regular arrangement, remove redundant and disordered data, identify and mark sequence anomalies through sliding window operation, and construct a standardized basic dataset. Taking cell #1 as the object, the time-series sequences of capacity efficiency degradation index are read: 30, 31, 30, 31, 31; and the time-series sequences of maintenance urgency index are read: 0.75, 0.76, 0.75, 0.75, 0.76. These are arranged one-to-one according to the same timestamp, and duplicate timestamps and redundant backup data are deleted. A sliding window of length 5 is used to traverse the sequence, and the window mean and standard deviation are calculated. There are no isolated outliers exceeding twice the standard deviation, and no outliers are marked. The data is then regularly arranged in chronological order to form a time-aligned and error-free standardized basic dataset, providing a unified input for subsequent verification. Secondly, a dual verification of temporal structure and continuity was conducted to distinguish different types of abnormal data. An adaptation method was used to replace and correct abnormal values. Basic anomaly screening was completed in conjunction with the operating conditions. It was checked that the collection interval of the two sets of sequences was fixed at 1 hour, with no structural errors such as temporal reversal, interruption, or missing data. The continuity of values ​​was checked. The fluctuation range of the capacity efficiency decay index was 0-100, and the fluctuation range of the maintenance urgency index was 0-1, both within a reasonable range, with no sudden increase, sudden decrease, or discontinuity. All values ​​in this window were under stable charging conditions, with no value jumps caused by sudden changes in operating conditions. All were determined to be valid data, requiring no replacement or correction. Basic anomaly screening was completed, and the original valid values ​​were retained.

[0060] Next, trend identification and autocorrelation analysis were performed on the indicator sequences to screen for abnormal fluctuations and lagged anomalies. Based on neighborhood verification and comparison with contemporaneous data, deep anomaly identification and numerical correction were completed. The capacity efficiency decay index showed stable and slight fluctuations, and the maintenance urgency index remained stable, both consistent with the slow and gradual trend of battery aging. Lagged 1, 2, and 3 autocorrelation analyses were performed on the two sets of sequences, and the correlation coefficients were all greater than 0.5, with no low-correlation lagged anomalies. Neighborhood verification was performed on each value at two time points before and after, and the proportion of normal values ​​in the neighborhood was all higher than 60%. There were no values ​​that deviated significantly from the normal values ​​of the same period, no deep anomalies, no need for correction, and the original purified data were retained. Finally, the time-series data of the two types of indicators are purified and hierarchical coding is performed separately. Abnormal disturbances in the coding sequences are removed. The two sets of codes are matched and combined one by one according to the time sequence to generate a standardized combined decision coding sequence. For the capacity efficiency decay indicator, the hierarchical coding is as follows: less than 10 is 0, 10–30 is 1, and greater than 30 is 2. All values ​​in this group are coded as 1. For the maintenance urgency index, the hierarchical coding is as follows: less than 0.3 is 0, 0.3–0.7 is 1, and greater than 0.7 is 2. All values ​​in this group are coded as 2. After removing disturbances without obvious abrupt changes in the coding sequence, each set of maintenance urgency codes and capacity efficiency codes are combined one by one according to the time sequence to generate a standardized 6-bit combined decision coding sequence: 21, 21, 21, 21, 21. This coding sequence is directly used for subsequent hierarchical decoding, achieving time-series alignment of indicators and anomaly screening. The coding is stable and reliable, the decisions are accurate and reproducible, and the maintenance level can be objectively output, improving operational efficiency and ensuring battery operation safety.

[0061] refer to Figure 2 A battery cabinet maintenance system based on fault prediction and health management includes: The multi-dimensional parameter processing module 100 is used to collect multi-dimensional operating parameters of the battery cabinet and store them according to the operating stage, calculate the parameter variation coefficient and dynamic weight, and obtain the stage health index by weighted summation of parameter deviation. Individual health assessment module 200 is used to obtain an individual health score based on the fusion of the phased health index, calculate and correct the health score by combining the historical out-of-limit frequency, and accumulate the deviation difference between the individual and the group median to obtain the historical deviation total value. The abnormal state identification module 300 is used to determine the dynamic boundary based on historical parameters, compare the current parameters with the dynamic boundary to obtain a comprehensive risk score, divide the time window to extract feature vectors to calculate the trajectory deviation, and identify abnormal transitions by combining the parameter change rate. The Population Deviation Analysis Module 400 is used to calculate the population deviation index in the dimensions of voltage, temperature and SOC to identify outlier subsets, and to obtain the capacity efficiency degradation index by statistically analyzing the capacity decay rate and charge and discharge efficiency. The maintenance decision output module 500 is used to obtain the maintenance urgency index by weighted summation of the historical deviation total value, comprehensive risk score, trajectory deviation degree, abnormal jump and outlier subset ratio, and output hierarchical maintenance decisions based on the maintenance urgency index and capacity efficiency decay index.

[0062] Reference Figure 3 This application also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, this computer device includes a processor, memory, network interface, and database connected via a bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores operations, computer programs, and the database. The internal memory provides an environment for the operation of the operations and computer programs stored in the non-volatile storage media. The database stores data such as battery cabinet maintenance methods based on fault prediction and health management. The network interface is used for communication with external terminals via a network connection. When executed by a processor, this computer program implements a battery cabinet maintenance method based on fault prediction and health management, including the following steps: collecting multi-dimensional operating parameters of the battery cabinet and storing them according to operating stages; calculating the parameter variation coefficient and dynamic weight; and weighted summing the parameters based on the parameter deviation to obtain a staged health index; fusing the staged health index to obtain an individual health score; calculating a corrected health score based on historical over-limit frequencies; accumulating the deviation difference between the individual and the group median to obtain the historical deviation total; determining the dynamic boundary based on historical parameters; comparing the current parameters with the dynamic boundary to obtain a comprehensive risk score; dividing the time window to extract feature vectors and calculate the trajectory deviation; and identifying abnormal transitions based on the parameter change rate; calculating the group deviation index using voltage, temperature, and SOC as dimensions to identify outliers; and statistically analyzing the capacity decay rate and charging / discharging efficiency to obtain a capacity efficiency decay index; weighted summing the historical deviation total, comprehensive risk score, trajectory deviation, abnormal transitions, and outlier percentage to obtain a maintenance urgency index; and outputting a graded maintenance decision based on the maintenance urgency index and the capacity efficiency decay index.

[0063] One embodiment of this application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements a battery cabinet maintenance method based on fault prediction and health management, including the following steps: collecting multi-dimensional operating parameters of the battery cabinet and storing them according to operating stages; calculating the parameter variation coefficient and dynamic weight; and weighted summing the parameters based on the parameter deviation to obtain a staged health index; fusing the staged health index to obtain a single-unit health score; calculating a corrected health score based on historical over-limit frequencies; accumulating the deviation difference between the single-unit and the group median to obtain a historical deviation total value; determining the dynamic boundary based on historical parameters; comparing the current parameters with the dynamic boundary to obtain a comprehensive risk score; dividing the time window to extract feature vectors to calculate the trajectory deviation; and identifying abnormal transitions based on the parameter change rate; calculating the group deviation index using voltage, temperature, and SOC as dimensions to identify outliers; and statistically analyzing the capacity decay rate and charging / discharging efficiency to obtain a capacity efficiency decay index; weighted summing the historical deviation total value, comprehensive risk score, trajectory deviation, abnormal transitions, and outlier percentage to obtain a maintenance urgency index; and outputting a graded maintenance decision based on the maintenance urgency index and the capacity efficiency decay index.

[0064] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0065] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A battery cabinet maintenance method based on fault prediction and health management, characterized in that, include: Collect multidimensional operating parameters of the battery cabinet and store them according to the operating stage. Calculate the parameter variation coefficient and dynamic weight, and obtain the staged health index by weighted summation of parameter deviation. The individual health score is obtained by fusing the phased health index, and the corrected health score is calculated by combining the historical out-of-limit frequency. The total historical deviation value is obtained by accumulating the deviation difference between the individual and the group median. The dynamic boundary is determined based on historical parameters. The current parameters are compared with the dynamic boundary to obtain a comprehensive risk score. The time window is divided to extract feature vectors to calculate the trajectory deviation. Abnormal transitions are identified by combining the parameter change rate. The population deviation index is calculated using voltage, temperature and SOC as dimensions to identify outlier subsets, and the capacity efficiency degradation index is obtained by statistically analyzing the capacity decay rate and charge / discharge efficiency. The maintenance urgency index is obtained by weighted summing of the historical deviation total, comprehensive risk score, trajectory deviation degree, abnormal jump and outlier subset proportion. Based on the maintenance urgency index and capacity efficiency decay index, a graded maintenance decision is output.

2. The battery cabinet maintenance method based on fault prediction and health management according to claim 1, characterized in that, The steps of collecting multi-dimensional operating parameters of the battery cabinet and storing them according to operating stages, calculating the parameter variation coefficient and dynamic weight, and obtaining the staged health index by weighted summation of parameter deviations include: Collect multi-dimensional operating parameters of the battery cabinet, divide the time sequence into different operating stages and form a stage parameter matrix, fill in missing data, and complete parameter classification and storage. The coefficient of variation is calculated for the parameter matrix of each stage. The weighted coefficient of variation is obtained by introducing the time decay factor. The comprehensive coefficient of variation is extracted and normalized. The dynamic weight is determined after secondary correction and weight compensation. The parameter deviation is calculated based on the stage statistical benchmark, and the deviation is truncated and fused. The final deviation is obtained by combining the parameter change information, and the abnormal frequency is fed back to the weight iteration. Multiply the dynamic weights by the final absolute value of the deviation and sum them up to obtain the phased health index for the current stage. Store the health indexes by stage and establish a time series sequence. Output the health index values ​​with stage labels.

3. The battery cabinet maintenance method based on fault prediction and health management according to claim 2, characterized in that, The steps of calculating the coefficient of variation for the parameter matrix at each stage, introducing a time decay factor to obtain a weighted coefficient of variation, extracting and normalizing the comprehensive coefficient of variation, and determining the dynamic weights after secondary correction and weight compensation include: Read the parameter matrix of each stage and sort it by time, set the corresponding time window, filter the valid data segment and remove abnormal data, and organize it into a standard parameter sequence according to the timestamp. The basic coefficient of variation is calculated for the standard parameter sequence, the time decay coefficient is assigned and the weighted statistic is calculated to obtain the weighted coefficient of variation, and the parameter covariance matrix is ​​extracted to obtain the comprehensive variability index. The statistical parameters are used to generate a comprehensive correction coefficient based on the abnormal correlation characteristics. The weighted coefficient of variation is then double-corrected by combining the degree of variation index. Finally, a lower limit constraint on the weights is applied to obtain the intermediate weight value. The intermediate weight values ​​are normalized, and the abnormally frequent parameters are compensated. The compensated weights are normalized again, and the final dynamic weights are output and the weight iteration records are stored.

4. The battery cabinet maintenance method based on fault prediction and health management according to claim 1, characterized in that, The steps of obtaining an individual health score based on the phased health index fusion, calculating a corrected health score by combining historical exceedance frequencies, and accumulating the deviation difference between the individual and the population median to obtain the total historical deviation include: Read the health index of each operation stage and perform the corresponding numerical transformation. Assign weights according to the operation stage and perform weighted fusion. After removing abnormal extreme values, obtain the individual health score. The historical parameters of individual units are statistically analyzed to exceed the limits. The frequency of each item exceeding the limit is calculated by hierarchical weighting and introducing a time decay factor. The weighted fusion is used to obtain the comprehensive frequency of exceeding the limits, and the corrected health score is obtained by calculating it with the individual health score. Extract and sort the population modified health scores to determine the population median, calculate the absolute difference between the individual and the median, apply an adjustment coefficient based on the historical deviation level, and obtain the adjusted deviation difference; The adjusted deviation difference is added to the historical cumulative deviation value, weighted according to time and accumulated. Constraint processing is applied to the cumulative result to form and output the historical total deviation value with individual identifiers.

5. The battery cabinet maintenance method based on fault prediction and health management according to claim 1, characterized in that, The steps of determining the dynamic boundary based on historical parameters, comparing the current parameters with the dynamic boundary to obtain a comprehensive risk score, dividing the time window to extract feature vectors to calculate the trajectory deviation, and combining the parameter change rate to identify abnormal transitions include: Historical parameter data is read and multi-level quantiles are calculated according to time period. A dynamic boundary matrix is ​​constructed, and the current parameters are compared with the dynamic boundary to obtain the comprehensive risk score. The time window is divided based on the charge-discharge cycle. A window sequence is generated by sliding sampling. The statistical features of parameters within the window are extracted and combined. After standardization, a multidimensional feature vector is obtained. Construct a historical normal window baseline feature set, calculate the multi-class distance between the current feature vector and the baseline set, obtain the trajectory deviation after averaging and normalization, and mark abnormal windows based on dynamic thresholds; The parameter change rate is calculated point by point and the operation level is divided. The jumps and abnormal oscillations between levels are detected. The abnormal transition judgment is completed by combining the trajectory deviation and the relevant transition information is recorded.

6. The battery cabinet maintenance method based on fault prediction and health management according to claim 5, characterized in that, The steps of dividing the time window based on the charge-discharge cycle, generating a window sequence using sliding sampling, extracting and combining statistical features of parameters within the window, and obtaining a multi-dimensional feature vector through standardization include: A single operating condition time window is divided by the charge and discharge cycle. A window sequence is generated by sliding sampling. Time series data of core parameters within the window are collected, and valid data is filtered and abnormal invalid data is removed. Calculate basic statistics for parameters within the window, perform frequency domain transformation on the basic statistics, generate corresponding derived statistics, and form a complete set of parameter features; The basic and derived statistics are combined hierarchically according to fixed dimensions to construct the original feature vector. The original feature vector is then subjected to dimensional unification processing to unify the vector format. The original feature vector is standardized according to dimensional statistics, the standardized values ​​are symmetrically truncated, and protection processing is completed by combining hierarchical labels, outputting a standardized multidimensional feature vector with window labels.

7. The battery cabinet maintenance method based on fault prediction and health management according to claim 1, characterized in that, The steps of calculating the population deviation index using voltage, temperature, and SOC as dimensions to identify outlier subsets, and statistically analyzing the capacity decay rate and charge / discharge efficiency to obtain the capacity efficiency decay index, include: The distance between individual units is calculated using voltage, temperature, and SOC as dimensions. The group deviation index is determined based on the distance, and outlier subsets are identified based on the index. Collect the rated capacity and charge / discharge cycle data of individual cells, remove invalid samples, calculate the adjacent capacity decay rate and perform sign and constraint processing, and sum them to obtain the cumulative capacity decay rate. Calculate the total cumulative capacity decay and average decay level of individual cells, calculate the single charge-discharge efficiency and make interval correction, and determine the steady-state charge-discharge efficiency based on effective cycle data. Historical data on capacity decay rate and charge / discharge efficiency are extracted and relevant statistics are calculated. After standardization and nonlinear fusion, and then exponential transformation and percentage processing, the capacity efficiency decay index is obtained and output.

8. The battery cabinet maintenance method based on fault prediction and health management according to claim 7, characterized in that, The steps of collecting the rated capacity and charge / discharge cycle data of individual cells, removing invalid samples, calculating adjacent capacity decay rates and performing sign and constraint processing, and summing them to obtain the cumulative capacity decay rate include: Collect charge-discharge cycle data of individual battery cells throughout their entire life cycle, verify the integrity of the data and the rationality of the operating conditions, remove invalid cycle samples, select valid cycles and organize them in an orderly manner to construct a continuous capacity dataset. Using adjacent effective cycles as the calculation unit and the capacity of the preceding cycle as the benchmark, a standardized adjacent capacity decay rate is obtained, and the decay rate is corrected for the aging stage. The signs of the corrected adjacent capacity decay rates are processed, the negative decay data corresponding to irreversible aging is retained, the positive fluctuation data is set to zero, and then the corresponding weights are assigned according to time and accumulated in a weighted manner. The single-cycle decay rate after successive accumulation is processed, and the monotonicity and boundary constraints of the cumulative result are applied. After normalization, the cumulative capacity decay rate with aging indicator is output.

9. The battery cabinet maintenance method based on fault prediction and health management according to claim 1, characterized in that, The steps of obtaining a maintenance urgency index by weighted summation of the historical deviation total, comprehensive risk score, trajectory deviation degree, abnormal jumps, and outlier subset proportion, and outputting tiered maintenance decisions based on the maintenance urgency index and capacity efficiency decay index, include: Obtain relevant evaluation indicators such as historical total deviation and comprehensive risk score, classify and encode each indicator, combine them to form maintenance urgency code, and set graded judgment boundaries in combination with equipment operation attributes; Obtain capacity efficiency decay indicators, classify and encode them, integrate the two types of codes to generate combined decision codes, conduct validity verification of relevant indicators, and screen out abnormal values; The combined decision code is decoded in a hierarchical manner, and the risk level is maintained according to the coding level. Boundary correction and verification are completed by combining the attenuation index. The decoded maintenance decisions are smoothed, matched with preset standardized maintenance plans, the maintenance operation requirements are clarified, and the hierarchical maintenance decision content is output.

10. A battery cabinet maintenance system based on fault prediction and health management, characterized in that, include: The multi-dimensional parameter processing module is used to collect multi-dimensional operating parameters of the battery cabinet and store them according to the operating stage, calculate the parameter variation coefficient and dynamic weight, and obtain the stage health index by weighted summation of parameter deviation. The individual health assessment module is used to obtain an individual health score based on the fusion of the phased health index, calculate and correct the health score by combining the historical out-of-limit frequency, and accumulate the deviation difference between the individual and the group median to obtain the historical deviation total value. The abnormal state identification module is used to determine the dynamic boundary based on historical parameters, compare the current parameters with the dynamic boundary to obtain a comprehensive risk score, divide the time window to extract feature vectors to calculate the trajectory deviation, and identify abnormal transitions by combining the parameter change rate. The group deviation analysis module is used to calculate the group deviation index in terms of voltage, temperature and SOC to identify outlier subsets, and to obtain the capacity efficiency degradation index by statistically analyzing the capacity decay rate and charge and discharge efficiency. The maintenance decision output module is used to obtain the maintenance urgency index by weighted summation of the historical total deviation value, comprehensive risk score, trajectory deviation degree, abnormal jump and outlier subset ratio, and output hierarchical maintenance decisions based on the maintenance urgency index and capacity efficiency decay index.

Images